Method, device and equipment for generating search space and storage medium

Abstract

The invention discloses a method, device and equipment for generating a search space and a storage medium, and relates to the field of artificial intelligence such as deep learning and computer vision. According to the specific implementation scheme, the method comprises the steps of: expanding an initial search space; searching a network structure unit in the initial search space, generating an initial model structure according to the searched network structure unit and the structured information of the target model, and training the initial model structure to record the performance of the candidate model obtained by training; determining an optimal search space according to the performance of the alternative model, searching an optimal target model based on the optimal search space, andtherefore, the precision and efficiency of the target model applied to multiple fields are improved, and the precision and the efficiency based on the same hardware target model are higher. On the premise of ensuring the same precision and efficiency, the method can be implemented by using cheaper hardware, and the hardware cost is reduced.

Description

technical field [0001] This application relates to the field of artificial intelligence, specifically deep learning and computer vision, and in particular to a method, device, device and storage medium for generating a search space. Background technique [0002] In recent years, deep learning technology has achieved great success in many directions. In deep learning technology, the quality of the artificial neural network structure has a very important impact on the effect of the final model. Manually designing network topology requires a lot of experience and many attempts, and many parameters will produce explosive combinations. Conventional random search is almost infeasible, so Neural Architecture Search (Neural Architecture Search, NAS for short) has become a research hotspot. [0003] In NAS, the search space is very important. The search space in the existing NAS is artificially designed. Given a small number of possible model structures, the number of search channels...

AI Technical Summary

Problems solved by technology

[0003] In NAS, the search space is very important. The search space in the existing NAS is artificially designed. Given a small number of possible model structures, the number of search channels, expansion coefficient, etc., there are great limitations. Searching for the optimal model structure among a small number of possible model structures in the search space, the performance of the finally found model structure is poor, and the accuracy and efficiency of image processing, natural language processing, audio / video processing and other data processing are relatively low. Low

Images